Your data and security
How AMOS keeps your company's data isolated, your credentials server-side, and consequential actions under human control.
AMOS is the single governed path between an AI and your company's systems. Identity, company boundaries, credentials, policy, approvals and proof all stay in one control plane.
Your data stays in your company
- The signed-in connection fixes which company you are working in. No tool argument can select a different company.
- Role and scope checks limit every user and API key. Protected company tables use PostgreSQL row-level security.
- Higher plans add dedicated data (Pro) and a dedicated VPC (Compliance).
Credentials stay server-side
- Connection secrets and OAuth tokens remain in the AMOS Platform vault, encrypted with ChaCha20-Poly1305.
- They are never returned to AMOS Desktop or to the AI model.
- Connector calls are blocked from reaching loopback, private, link-local and cloud-metadata addresses.
People set the ceiling, not the AI
- Built-in owner, admin, member and viewer roles set what each person can do. Owners can narrow an individual's access further.
- An API key can never do more than the person who issued it.
- The AI cannot grant itself a role, widen a connection's scope, or approve its own work.
Consequential actions wait for a person
AMOS tells a read apart from a draft, a publication, a payment, an outgoing message or a deletion. Consequential actions can wait for a signed-in person to approve them, and actions AMOS cannot classify safely are refused rather than allowed.
Proof after every action
Governed actions create durable receipts that link who acted, the company, the policy, the inputs, the result and the time, without exposing credentials.
Encryption
- HTTPS/TLS on public AMOS endpoints.
- Application-layer encryption for connection credentials.
- AES-256 server-side encryption for AMOS-managed object storage, with public access blocked.
More detail
See amoslabs.com/security, which also offers a two-page security and trust brief.